Systematic discovery of directional regulatory motifs associated with human insulator sites

Naoki Osato, Michiaki HamadabioRxiv 2024 · DOI

Abstract

Insulator proteins function as barriers to enhancer–promoter interactions (EPIs), thereby regulating gene expression. While CTCF is recognized as the primary insulator protein in vertebrates, the roles of other DNA-binding proteins (DBPs) in EPI insulation remain poorly understood. We developed a systematic deep learning–based approach to identify DNA motifs of DBPs associated with insulator function. Applying this method to human fibroblast cells, we identified 97 directional motifs mapped to 23 DBPs previously linked to insulator activity or chromosomal regulation. Our findings demonstrate that CTCF’s orientation bias is proportional to chromatin interaction data and that these motifs are significantly enriched at chromatin boundaries and splice sites. These results support the insulator-pairing model, suggesting that insulator–insulator interactions are orientation-dependent and involve a diverse array of proteins beyond CTCF.

Introduction

In the complex three-dimensional environment of the human nucleus, precise gene regulation depends on the ability of enhancers to find their target promoters while avoiding inappropriate cross-talk. This spatial organization is maintained by insulators—protein-DNA complexes that act as physical barriers or anchors for chromatin loops.

For decades, the protein CTCF has been considered the "master weaver" of the genome due to its unique directional bias; it typically forms loops only when two binding sites are in a convergent (Forward-Reverse) orientation. However, recent high-resolution mapping (Micro-C) has revealed many genomic loops that exist independently of CTCF. Identifying these "hidden" insulators experimentally is a laborious process. This study addresses the challenge by utilizing a computational framework to systematically discover DBPs whose orientation and position influence gene expression and chromatin architecture.

Methodology

The researchers adapted a deep learning tool called DEcode, originally designed to predict gene expression from promoter sequences, to identify insulator-associated motifs.

Model Architecture and Data Integration

The core of the approach is a 1D convolutional neural network (CNN) that processes ChIP-seq peak data (representing protein-binding events) to predict mRNA levels. Unlike traditional models that focus solely on the immediate vicinity of a gene, this framework integrated eQTL data and "nearest gene" association rules to incorporate distal enhancer regions up to 1 Megabase away. The study utilized a massive dataset, including RNA-seq from four human cell types, ChIP-seq data for 1,356 DBPs, and high-resolution Micro-C and ChIA-PET data for structural validation.

Identifying Directional Bias

To interpret the model's predictions, the team employed DeepLIFT, an algorithm that assigns "contribution scores" to specific binding sites. By comparing these scores across four orientations—Forward-Reverse (FR), Reverse-Forward (RF), Forward-Forward (FF), and Reverse-Reverse (RR)—the researchers could identify proteins whose regulatory impact changed significantly based on their physical orientation on the DNA. Proteins exhibiting this behavior were flagged as potential directional insulators.

Key Insights and Results

Discovery of 23 Directional DBPs

The model achieved a high predictive accuracy, with a Spearman’s correlation of 0.74 between predicted and actual gene expression. It successfully validated the known behavior of CTCF, estimating an orientation bias (76.4% in favor of FR) that matched experimental ChIA-PET data. Beyond CTCF, the study identified 23 DBPs—including YY1, ZNF143, and MAZ—that exhibit significant directional bias associated with insulator functions.

The "Barrier" Function and Chromatin Borders

A defining characteristic of insulators is their ability to separate distinct chromatin domains. The identified motifs were frequently located at the exact borders between active (H3K4me3-rich) and repressed (H3K27me3-rich) chromatin. Specifically, predicted insulator sites showed a greater than two-fold difference in repressive marks between their upstream and downstream regions compared to non-insulator sites, suggesting they act as "firewalls" against the spread of silencing signals.

Tissue-Specific Regulation

The study highlighted the role of these insulators in tissue-specific functions. For example, the MyoD binding site was identified as an insulator near genes critical for skeletal muscle differentiation. Furthermore, these proteins appear to operate independently of the CTCF-cohesin machinery, suggesting the existence of multiple, parallel insulation systems within the genome.

Discussion and Future Directions

These findings provide strong computational evidence for the insulator-pairing model, which posits that genomic architecture is governed by the specific orientation of various DNA-binding proteins. By identifying 23 proteins that function alongside or independently of CTCF, this research expands our understanding of the "genomic grammar" that dictates 3D structure.

While the study is limited by the availability of high-quality ChIP-seq data for less common DBPs, it provides a scalable pipeline for genomic discovery. Future work may involve using these insights to design synthetic insulators for gene therapy, allowing for the precise shielding of therapeutic genes from their surrounding genomic environment. Experimental validation, such as CRISPR-mediated motif inversion, remains the next step in confirming the causal role of these newly identified directional regulators.

Summary updated

Abstract

Abstract Insulator proteins function as barriers to enhancer–promoter interactions (EPIs), thereby regulating gene expression. The primary insulator protein in vertebrates is CTCF, a DNA-binding protein (DBP); however, the roles of other DBPs in EPI insulation are not fully understood. To address this, we developed a systematic and comprehensive deep learning–based approach to identify DNA motifs of DBPs associated with insulator function. Applying this method to human fibroblast cells, we identified 97 directional motifs and a smaller number of non-directional motifs. These motifs were mapped to 23 DBPs previously linked to insulator activity, CTCF, and/or other forms of chromosomal transcriptional regulation. We found that the estimated orientation bias of CTCF was consistently proportional to the orientation bias observed in chromatin interaction data. Furthermore, these motifs showed significant enrichment at insulator sites that separate repressive and active chromatin regions, at chromatin interaction–defined boundaries, and at splice sites, compared to motifs of other DBPs. For instance, we observed that the key regulator MyoD-binding site is located at an insulator site near a gene involved in skeletal muscle differentiation and function. Importantly, our findings support the previously proposed insulator-pairing model, which suggests that insulator–insulator interactions are orientation-dependent, and highlight the involvement of multiple DNA-binding proteins beyond CTCF. Together, these results provide new insights into transcriptional regulatory mechanisms mediated by insulator-associated DBPs.
BibTeX
@misc{osato2024systematic,
  author = {Naoki Osato and Michiaki Hamada},
  title = {{Systematic discovery of directional regulatory motifs associated with human insulator sites}},
  year = {2024},
  doi = {10.1101/2024.01.20.573595},
  howpublished = {bioRxiv},
}
Figure